Research On Optimal Operation And Control Integration Of Process System

The manufacturing industry is the pillar industry of Chinese economic and social development,and the process industry represented by the chemical industry is an important part of it.At present,the production equipment in the Chinese process industry is becoming larger,the production capacity and output are expanding,the process chain and supply chain are becoming more complex,and the products and intermediate products are increasing.The problems faced by automatic systems are more complex and changeable.They put forward more complex and more challenging requirements for advanced process control and optimization in the process industry.To this end,this thesis focuses on the two core issues of the variable selection and operation optimization and control integration,several case studies are presented to illustrate the effectiveness of the proposed strategies.The main contributions of this thesis are summarized as follows:1)The concept of self-optimizing control(SOC)is introduced for the controlled variables(CVs)selection.A dynamic SOC strategy based on the null space method is proposed.We assume the measurement noise can be neglected,and the CVs can be represented as linear combinations of the measurement vector,where the combination matrix is easily obtained from optimal sensitivity analysis.The selected CVs are optimally invariant to disturbances,which avoids the "re-optimization" and reduces the frequency of the set-points update.Since the proposed strategy belongs to the dynamic optimization category,both the combination matrix and the optimal sensitivity matrix are time-varying,but they can be obtained through finite difference method,thereby reducing the computational burden.2)Considering the existing local SOC(linear SOC)method applied to nonlinear processes,the CVs only valid near the nominal operating point.A global SOC strategy based on the offline process model and simulation data is proposed.The nonlinear model is employed to calculate the average economic loss over the entire operating space in the presence of various disturbances and measurement noise.The proposed strategy is essentially a nonlinear programming problem.Reasonable assumptions are made for some conditions,and the entire problem is transformed into a convex optimization problem,the analytical expression of the CVs is obtained.Besides,mixed integer constraints are incorporated into the global SOC strategy with the aim of balancing the sensor investment and control system performance.The proposed method can handle additional structural constraints as well as determine the optimal subset of measurements with globally valid CVs.3)The hierarchy model of process system is analyzed,and the cascade structure of real-time optimization and control integration is proposed,aiming at the optimal operation problem for industrial process.Gradient information-based steady state real time optimization approach is installed in the optimization layer.The set-points are updated by collecting measurements online and estimating the gradient information of process.The proposed approach can effectively suppress the impact of plant model mismatch on optimization objective,since it avoids using an explicit process model.A least square technique is introduced to compute the gradient vector.The proposed algorithm not only has a low computational burden,but also can be applied to the steady state real time optimization of large-scale industrial processes.Furthermore,an eigenvalue decomposition-based method for selecting CVs of process is discussed.The proposed method minimizes the global average loss based on the nonlinear model Reasonable simplifications are made for some conditions,so that the suboptimal solution is obtained,in order to solve the nonlinear programming problem efficiently.The analytical solution of CVs is given and the calculation efficiency is improved as well as the optimization layer is connected with the control layer.4)To achieve the optimal operation and control integration for industrial process in the presence of disturbances and uncertainty,a novel hierarchical architecture combining extremum-seeking(ES)and SOC is proposed.The proposed architecture features two main components.The first is a data-driven fast ES approach that is employed in the upper optimization layer.The data-driven approach does not rely on the process model,which can effectively suppress the influence of unmodeled disturbances.In addition,compared with the traditional ES method,incorporating the plant dynamics during the gradient estimation can improve the convergence rate significantly.The second is to present a global SOC strategy based on the necessary conditions of optimality,which can effectively combine the optimization layer with control layer,and make a fast adjustment to expected disturbances.The proposed strategy does not require the second order derivative information,therefore,it is numerically more reliable and robust.